UPAD: A Large-Scale Passive Sonar Benchmark Dataset for Vessel Detection and Classification

In the realm of underwater acoustics, the complex and dynamic environment poses formidable challenges for the detection and classification of vessels using passive sonar systems. Recent strides in deep learning (DL) have sparked optimism in automating or enhancing the data analysis process, traditionally reliant on human expertise. However, the efficacy of DL hinges on substantial training data, a resource currently scarce in the public domain, particularly for annotated real-world passive sonar data. This study endeavors to bridge this gap by presenting a methodology to construct a sizable annotated dataset by leveraging automatic identification system (AIS) data. The outcome of this effort is the Underwater Passive Acoustic Dataset (UPAD), an extensive benchmark dataset meticulously crafted for vessel detection and multi-label classification using passive sonar. UPAD not only addresses the dearth of publicly available benchmark data but also facilitates advancements in the application of DL for navigating the complexities of underwater acoustic environments.